diff --git a/include/cufinufft/common.h b/include/cufinufft/common.h index efa7eb7b1..e1149df1f 100644 --- a/include/cufinufft/common.h +++ b/include/cufinufft/common.h @@ -41,8 +41,8 @@ template void cufinufft_setup_binsize(int type, int ns, int dim, cufinufft_opts *opts); template -auto cufinufft_set_shared_memory(V *kernel, const int dim, - const cufinufft_plan_t &d_plan) { +int cufinufft_set_shared_memory(V *kernel, const int dim, + const cufinufft_plan_t &d_plan) { /** * WARNING: this function does not handle cuda errors. The caller should check them. */ diff --git a/include/finufft.h b/include/finufft.h index 487a3eb4f..18eab5921 100644 --- a/include/finufft.h +++ b/include/finufft.h @@ -35,10 +35,6 @@ #include #define FINUFFT_BIGINT int64_t -#ifndef __cplusplus -#include // for bool type in C (needed for item in plan struct) -#endif - // this macro name has to be safe since exposed to user #define FINUFFT_SINGLE #include diff --git a/include/finufft/defs.h b/include/finufft/defs.h index 7184bdda8..c8ef5a116 100644 --- a/include/finufft/defs.h +++ b/include/finufft/defs.h @@ -24,18 +24,16 @@ // All indexing in library that potentially can exceed 2^31 uses 64-bit signed. // This includes all calling arguments (eg M,N) that could be huge someday. -#define BIGINT int64_t -#define UBIGINT uint64_t +using BIGINT = int64_t; +using UBIGINT = uint64_t; // Precision-independent real and complex types, for private lib/test compile #ifdef SINGLE -#define FLT float +using FLT = float; #else -#define FLT double +using FLT = double; #endif -// next line possibly obsolete... -#define _USE_MATH_DEFINES #include // we define C++ complex type only -#define CPX std::complex +using CPX = std::complex; // inline macro, to force inlining of small functions // this avoids the use of macros to implement functions @@ -65,44 +63,49 @@ // ------------- Library-wide algorithm parameter settings ---------------- // Library version (is a string) -#define FINUFFT_VER "2.3.0" +#define FINUFFT_VER "2.3.0" // Smallest possible kernel spread width per dimension, in fine grid points // (used only in spreadinterp.cpp) -#define MIN_NSPREAD 2 +inline constexpr int MIN_NSPREAD = 2; // Largest possible kernel spread width per dimension, in fine grid points // (used only in spreadinterp.cpp) -#define MAX_NSPREAD 16 +inline constexpr int MAX_NSPREAD = 16; // Fraction growth cut-off in utils:arraywidcen, sets when translate in type-3 -#define ARRAYWIDCEN_GROWFRAC 0.1 +inline constexpr double ARRAYWIDCEN_GROWFRAC = 0.1; // Max number of positive quadr nodes for kernel FT (used only in common.cpp) -#define MAX_NQUAD 100 +inline constexpr int MAX_NQUAD = 100; // Internal (nf1 etc) array allocation size that immediately raises error. // (Note: next235 takes 1s for 1e11, so it is also to prevent hang here.) // Increase this if you need >10TB (!) RAM... -#define MAX_NF (BIGINT)1e12 +inline constexpr BIGINT MAX_NF = BIGINT(1e12); // Maximum allowed number M of NU points; useful to catch incorrectly cast int32 // values for M = nj (also nk in type 3)... -#define MAX_NU_PTS (BIGINT)1e14 +inline constexpr BIGINT MAX_NU_PTS = BIGINT(1e14); // -------------- Math consts (not in math.h) and useful math macros ---------- #include // either-precision unit imaginary number... #define IMA (CPX(0.0, 1.0)) -// using namespace std::complex_literals; // needs C++14, provides 1i, 1if + +// MR: In the longer term I suggest to move +// away from M_PI, which was never part of the standard. +// Perhaps a constexpr pi in the namespace finufft, or a constexpr finufft_pi +// if no namespaces are used? +// In C++20 these constants will be part of the language, and the problem will go away. #ifndef M_PI // Windows apparently doesn't have this const #define M_PI 3.14159265358979329 #endif #define M_1_2PI 0.159154943091895336 #define M_2PI 6.28318530717958648 // to avoid mixed precision operators in eg i*pi, an either-prec PI... -#define PI (FLT) M_PI +#define PI FLT(M_PI) // machine epsilon for decisions of achievable tolerance... #ifdef SINGLE @@ -115,19 +118,28 @@ // These macros should probably be replaced by modern C++ std lib or random123. // (RAND_MAX is in stdlib.h) #include -// #define rand01() (((FLT)(rand()%RAND_MAX))/RAND_MAX) -#define rand01() ((FLT)rand() / (FLT)RAND_MAX) +static inline FLT rand01() { return FLT(rand()) / FLT(RAND_MAX); } // unif[-1,1]: -#define randm11() (2 * rand01() - (FLT)1.0) +static inline FLT randm11() { return 2 * rand01() - FLT(1); } // complex unif[-1,1] for Re and Im: -#define crandm11() (randm11() + IMA * randm11()) +static inline CPX crandm11() { return randm11() + IMA * randm11(); } // Thread-safe seed-carrying versions of above (x is ptr to seed)... +// MR: we have to leave those as macros for now, as "rand_r" is deprecated +// and apparently no longer available on Windows. +#if 1 #define rand01r(x) ((FLT)rand_r(x) / (FLT)RAND_MAX) // unif[-1,1]: #define randm11r(x) (2 * rand01r(x) - (FLT)1.0) // complex unif[-1,1] for Re and Im: #define crandm11r(x) (randm11r(x) + IMA * randm11r(x)) +#else +static inline FLT rand01r(unsigned int *x) { return FLT(rand_r(x)) / FLT(RAND_MAX); } +// unif[-1,1]: +static inline FLT randm11r(unsigned int *x) { return 2 * rand01r(x) - FLT(1); } +// complex unif[-1,1] for Re and Im: +static inline CPX crandm11r(unsigned int *x) { return randm11r(x) + IMA * randm11r(x); } +#endif // ----- OpenMP macros which also work when omp not present ----- // Allows compile-time switch off of openmp, so compilation without any openmp @@ -135,16 +147,16 @@ #ifdef _OPENMP #include // point to actual omp utils -#define MY_OMP_GET_NUM_THREADS() omp_get_num_threads() -#define MY_OMP_GET_MAX_THREADS() omp_get_max_threads() -#define MY_OMP_GET_THREAD_NUM() omp_get_thread_num() -#define MY_OMP_SET_NUM_THREADS(x) omp_set_num_threads(x) +static inline int MY_OMP_GET_NUM_THREADS() { return omp_get_num_threads(); } +static inline int MY_OMP_GET_MAX_THREADS() { return omp_get_max_threads(); } +static inline int MY_OMP_GET_THREAD_NUM() { return omp_get_thread_num(); } +static inline void MY_OMP_SET_NUM_THREADS(int x) { omp_set_num_threads(x); } #else // non-omp safe dummy versions of omp utils... -#define MY_OMP_GET_NUM_THREADS() 1 -#define MY_OMP_GET_MAX_THREADS() 1 -#define MY_OMP_GET_THREAD_NUM() 0 -#define MY_OMP_SET_NUM_THREADS(x) +static inline int MY_OMP_GET_NUM_THREADS() { return 1; } +static inline int MY_OMP_GET_MAX_THREADS() { return 1; } +static inline int MY_OMP_GET_THREAD_NUM() { return 0; } +static inline void MY_OMP_SET_NUM_THREADS(int) {} #endif // Prec-switching name macros (respond to SINGLE), used in lib & test sources @@ -194,12 +206,11 @@ #include // (must come after complex.h) // group together a bunch of type 3 rescaling/centering/phasing parameters: -#define TYPE3PARAMS FINUFFTIFY(_type3Params) -typedef struct { - FLT X1, C1, D1, h1, gam1; // x dim: X=halfwid C=center D=freqcen h,gam=rescale - FLT X2, C2, D2, h2, gam2; // y - FLT X3, C3, D3, h3, gam3; // z -} TYPE3PARAMS; +template struct type3params { + T X1, C1, D1, h1, gam1; // x dim: X=halfwid C=center D=freqcen h,gam=rescale + T X2, C2, D2, h2, gam2; // y + T X3, C3, D3, h3, gam3; // z +}; typedef struct FINUFFT_PLAN_S { // the main plan object, fully C++ @@ -243,7 +254,7 @@ typedef struct FINUFFT_PLAN_S { // the main plan object, fully C++ CPX *deconv; // reciprocal of kernel FT, phase, all output NU pts CPX *CpBatch; // working array of prephased strengths FLT *Sp, *Tp, *Up; // internal primed targs (s'_k, etc), allocated - TYPE3PARAMS t3P; // groups together type 3 shift, scale, phase, parameters + type3params t3P; // groups together type 3 shift, scale, phase, parameters FINUFFT_PLAN innerT2plan; // ptr used for type 2 in step 2 of type 3 // other internal structs; each is C-compatible of course @@ -255,6 +266,4 @@ typedef struct FINUFFT_PLAN_S { // the main plan object, fully C++ } FINUFFT_PLAN_S; -#undef TYPE3PARAMS - #endif // DEFS_H diff --git a/include/finufft/fft.h b/include/finufft/fft.h index 7b0528303..906f62440 100644 --- a/include/finufft/fft.h +++ b/include/finufft/fft.h @@ -3,10 +3,10 @@ #ifdef FINUFFT_USE_DUCC0 #include "ducc0/fft/fftnd_impl.h" -#define FFTW_FORGET_WISDOM() // temporary hack since some tests call this unconditionally -#define FFTW_CLEANUP() // temporary hack since some tests call this unconditionally -#define FFTW_CLEANUP_THREADS() // temporary hack since some tests call this - // unconditionally +// temporary hacks to allow compilation of tests that assume FFTW is used +static inline void FFTW_FORGET_WISDOM() {} +static inline void FFTW_CLEANUP() {} +static inline void FFTW_CLEANUP_THREADS() {} #else #include "fftw_defs.h" #endif diff --git a/include/finufft/fftw_defs.h b/include/finufft/fftw_defs.h index 1771ff259..39f8d8c71 100644 --- a/include/finufft/fftw_defs.h +++ b/include/finufft/fftw_defs.h @@ -22,11 +22,7 @@ // now use this tool (note we replaced typedefs v<=2.0.4, in favor of macros): #define FFTW_CPX FFTWIFY(complex) #define FFTW_PLAN FFTWIFY(plan) -#define FFTW_ALLOC_RE FFTWIFY(alloc_real) #define FFTW_ALLOC_CPX FFTWIFY(alloc_complex) -#define FFTW_PLAN_1D FFTWIFY(plan_dft_1d) -#define FFTW_PLAN_2D FFTWIFY(plan_dft_2d) -#define FFTW_PLAN_3D FFTWIFY(plan_dft_3d) #define FFTW_PLAN_MANY_DFT FFTWIFY(plan_many_dft) #define FFTW_EX FFTWIFY(execute) #define FFTW_DE FFTWIFY(destroy_plan) diff --git a/include/finufft/spreadinterp.h b/include/finufft/spreadinterp.h index 173595593..78ecf9f22 100644 --- a/include/finufft/spreadinterp.h +++ b/include/finufft/spreadinterp.h @@ -20,10 +20,12 @@ NOTE: non-zero values are for experts only, since NUMERICAL OUTPUT MAY BE INCORRECT UNLESS finufft_spread_opts.flags=0 ! */ -#define TF_OMIT_WRITE_TO_GRID 1 // don't add subgrids to out grid (dir=1) -#define TF_OMIT_EVALUATE_KERNEL 2 // don't evaluate the kernel at all -#define TF_OMIT_EVALUATE_EXPONENTIAL 4 // omit exp() in kernel (kereval=0 only) -#define TF_OMIT_SPREADING 8 // don't interp/spread (dir=1: to subgrids) +enum { + TF_OMIT_WRITE_TO_GRID = 1, // don't add subgrids to out grid (dir=1) + TF_OMIT_EVALUATE_KERNEL = 2, // don't evaluate the kernel at all + TF_OMIT_EVALUATE_EXPONENTIAL = 4, // omit exp() in kernel (kereval=0 only) + TF_OMIT_SPREADING = 8 // don't interp/spread (dir=1: to subgrids) +}; namespace finufft { namespace spreadinterp { diff --git a/include/finufft_eitherprec.h b/include/finufft_eitherprec.h index d193a5922..47f7860e1 100644 --- a/include/finufft_eitherprec.h +++ b/include/finufft_eitherprec.h @@ -24,7 +24,6 @@ // decide which kind of complex numbers FINUFFT_CPX is (four options) #ifdef __cplusplus -#define _USE_MATH_DEFINES #include // C++ type #define FINUFFT_COMPLEXIFY(X) std::complex #else @@ -183,4 +182,3 @@ FINUFFT_EXPORT int FINUFFT_CDECL FINUFFTIFY(3d3many)( #undef FINUFFT_CPX #undef FINUFFT_PLAN #undef FINUFFT_PLAN_S -#undef FINUFFT_TYPE3PARAMS diff --git a/include/finufft_errors.h b/include/finufft_errors.h index 2feaa131a..245e6d339 100644 --- a/include/finufft_errors.h +++ b/include/finufft_errors.h @@ -3,24 +3,26 @@ // ---------- Global error/warning output codes for the library --------------- // All documentation is at ../docs/errors.rst (not here): -#define FINUFFT_WARN_EPS_TOO_SMALL 1 -#define FINUFFT_ERR_MAXNALLOC 2 -#define FINUFFT_ERR_SPREAD_BOX_SMALL 3 -#define FINUFFT_ERR_SPREAD_PTS_OUT_RANGE 4 // DEPRECATED -#define FINUFFT_ERR_SPREAD_ALLOC 5 -#define FINUFFT_ERR_SPREAD_DIR 6 -#define FINUFFT_ERR_UPSAMPFAC_TOO_SMALL 7 -#define FINUFFT_ERR_HORNER_WRONG_BETA 8 -#define FINUFFT_ERR_NTRANS_NOTVALID 9 -#define FINUFFT_ERR_TYPE_NOTVALID 10 -#define FINUFFT_ERR_ALLOC 11 -#define FINUFFT_ERR_DIM_NOTVALID 12 -#define FINUFFT_ERR_SPREAD_THREAD_NOTVALID 13 -#define FINUFFT_ERR_NDATA_NOTVALID 14 -#define FINUFFT_ERR_CUDA_FAILURE 15 -#define FINUFFT_ERR_PLAN_NOTVALID 16 -#define FINUFFT_ERR_METHOD_NOTVALID 17 -#define FINUFFT_ERR_BINSIZE_NOTVALID 18 -#define FINUFFT_ERR_INSUFFICIENT_SHMEM 19 -#define FINUFFT_ERR_NUM_NU_PTS_INVALID 20 +enum { + FINUFFT_WARN_EPS_TOO_SMALL = 1, + FINUFFT_ERR_MAXNALLOC = 2, + FINUFFT_ERR_SPREAD_BOX_SMALL = 3, + FINUFFT_ERR_SPREAD_PTS_OUT_RANGE = 4, // DEPRECATED + FINUFFT_ERR_SPREAD_ALLOC = 5, + FINUFFT_ERR_SPREAD_DIR = 6, + FINUFFT_ERR_UPSAMPFAC_TOO_SMALL = 7, + FINUFFT_ERR_HORNER_WRONG_BETA = 8, + FINUFFT_ERR_NTRANS_NOTVALID = 9, + FINUFFT_ERR_TYPE_NOTVALID = 10, + FINUFFT_ERR_ALLOC = 11, + FINUFFT_ERR_DIM_NOTVALID = 12, + FINUFFT_ERR_SPREAD_THREAD_NOTVALID = 13, + FINUFFT_ERR_NDATA_NOTVALID = 14, + FINUFFT_ERR_CUDA_FAILURE = 15, + FINUFFT_ERR_PLAN_NOTVALID = 16, + FINUFFT_ERR_METHOD_NOTVALID = 17, + FINUFFT_ERR_BINSIZE_NOTVALID = 18, + FINUFFT_ERR_INSUFFICIENT_SHMEM = 19, + FINUFFT_ERR_NUM_NU_PTS_INVALID = 20 +}; #endif diff --git a/src/finufft.cpp b/src/finufft.cpp index e88c484eb..ce4ada36d 100644 --- a/src/finufft.cpp +++ b/src/finufft.cpp @@ -91,13 +91,8 @@ namespace common { static std::mutex fftw_lock; #endif -// We macro because it has no FLT args but gets compiled for both prec's... -#ifdef SINGLE -#define SET_NF_TYPE12 set_nf_type12f -#else -#define SET_NF_TYPE12 set_nf_type12 -#endif -int SET_NF_TYPE12(BIGINT ms, finufft_opts opts, finufft_spread_opts spopts, BIGINT *nf) +static int set_nf_type12(BIGINT ms, finufft_opts opts, finufft_spread_opts spopts, + BIGINT *nf) // Type 1 & 2 recipe for how to set 1d size of upsampled array, nf, given opts // and requested number of Fourier modes ms. Returns 0 if success, else an // error code if nf was unreasonably big (& tell the world). @@ -686,16 +681,16 @@ int FINUFFT_MAKEPLAN(int type, int dim, BIGINT *n_modes, int iflag, int ntrans, } // determine fine grid sizes, sanity check.. - int nfier = SET_NF_TYPE12(p->ms, p->opts, p->spopts, &(p->nf1)); + int nfier = set_nf_type12(p->ms, p->opts, p->spopts, &(p->nf1)); if (nfier) return nfier; // nf too big; we're done p->phiHat1 = (FLT *)malloc(sizeof(FLT) * (p->nf1 / 2 + 1)); if (dim > 1) { - nfier = SET_NF_TYPE12(p->mt, p->opts, p->spopts, &(p->nf2)); + nfier = set_nf_type12(p->mt, p->opts, p->spopts, &(p->nf2)); if (nfier) return nfier; p->phiHat2 = (FLT *)malloc(sizeof(FLT) * (p->nf2 / 2 + 1)); } if (dim > 2) { - nfier = SET_NF_TYPE12(p->mu, p->opts, p->spopts, &(p->nf3)); + nfier = set_nf_type12(p->mu, p->opts, p->spopts, &(p->nf3)); if (nfier) return nfier; p->phiHat3 = (FLT *)malloc(sizeof(FLT) * (p->nf3 / 2 + 1)); } diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt index 37050876b..0e1a1730f 100644 --- a/test/CMakeLists.txt +++ b/test/CMakeLists.txt @@ -30,6 +30,7 @@ add_executable(testutils testutils.cpp) if(FINUFFT_USE_DUCC0) target_compile_definitions(testutils PRIVATE -DFINUFFT_USE_DUCC0) endif() +target_compile_features(testutils PRIVATE cxx_std_17) finufft_link_test(testutils) add_test( NAME run_testutils